Backdoor AI

Good morning.

Soon, yours truly is to be expected for a routine checkup at the hospital. Since bowel cancer is on the rise, they have lowered the age recommendation for colonoscopies from 50 to 45, and it’s expected they will lower that to 40.

Just like Cartman in the very first episode of South Park in 1997, I will be anally probed. Not by an alien but by a medical professional (i presume). I am NOT looking forward to that experience, to be honest.

I also hope I haven’t ruined your breakfast.

But it makes me wonder … there must be a better way than having someone shove a contraption up your backside.

You read a lot about medical mistakes in the news, and if there’s one field where AI will be extremely beneficial for humanity, it’s the medical arena.

From making correct diagnoses on scans and MRI’s to even finding cures to diseases,.

This week I saw there was a company called Endiatx that is developing an AI-driven robot in the form of a pill. As I understood it, you swallow the thing, and it goes on a journey through your digestive tract, taking pictures along the way, like a Japanese tourist.

In the first stages, it will be doctors guiding that thing; later on, with AI-powered guidance. I can see it now: a doctor with a little joystick driving a robot through your intestines. The privileges of 14 years of medical school.

I also wonder how you prevent that thing from going on a journey to the sea through the sewage system, but i’ll have to look deeper into that.

Anyway, , sorry again about your breakfast.

Welcome to the future (and to the Blacklynx Brief)

AI News

• Anthropic has launched Claude 3.5 Sonnet, an upgraded AI model that is twice as fast as its predecessor and significantly cheaper. It outperforms rivals like GPT-4o in various benchmarks including reasoning, coding, math, and knowledge. Additionally, Anthropic introduced a new feature called 'Artifacts' for real-time interaction, with more updates expected later this year.

  

  Source: Anthropic

• As mentioned in the intro : Endiatx's PillBot, a small swallowable robot, is undergoing clinical trials and preparing for FDA review. The device, about the size of a large vitamin pill, transmits high-resolution video to doctors in real-time, offering a less invasive alternative to traditional endoscopies and colonoscopies. Currently controlled by doctors, future versions of PillBot are expected to have AI-guided autonomous capabilities, with commercial availability targeted for 2026.

  

• Citigroup's new report predicts AI will significantly impact banking jobs, with over half of roles at high risk of automation. Banking is identified as the industry most susceptible to automation, followed by insurance and energy. The report also foresees AI adding $170 billion to the global banking sector's profits by 2028, with banks creating new roles for AI managers and compliance officers to navigate this transition.

• Apple is reportedly in discussions with Meta to incorporate Meta's generative AI models into Apple Intelligence, indicating a potential easing of their long-standing rivalry. Apple is considering partnerships with Meta, Perplexity, and Anthropic, alongside OpenAI, for its new AI system. This collaboration would allow these AI partners to offer premium subscriptions on Apple devices, with Apple taking a share of the revenue. This partnership would diversify Apple's AI offerings and provide Meta with a significant new distribution channel.

• Researchers have developed an AI-powered blood test that can predict Parkinson's disease up to 7 years before symptoms appear. The AI analyzes blood markers related to inflammation and protein degradation, identifying 79% of patients who later developed Parkinson's in a 10-year study. Researchers aim to create a simple finger-prick test for widespread screenings, potentially allowing earlier treatment and prevention of the disease.

• Universal Music Group, Warner Music Group, and Sony Music Entertainment have filed lawsuits against music generation startups Udio and Suno for allegedly training their AI models on copyrighted material without permission. The Recording Industry Association of America (RIAA) is leading the lawsuits, seeking up to $150,000 per infringed work, potentially amounting to billions in damages. The RIAA claims that the startups' AI outputs mimic specific artists and songs, and that the companies admitted to using copyrighted material for training. This case could set important precedents for the AI music industry or lead to significant settlements and licensing deals.

• OpenAI has acquired Multi, a startup specializing in remote collaboration and screensharing technology for Mac computers. Multi's team will join OpenAI’s ChatGPT desktop team, and their standalone platform will be discontinued. This acquisition, along with OpenAI's recent purchase of Rockset, suggests OpenAI aims to expand ChatGPT's integration into users' computers beyond simple chatbot functions

• OpenAI has delayed the launch of ChatGPT's Voice Mode feature, initially expected in May, to allow for further safety tests and refinements. The Alpha test group, planned for late June, has been postponed by a month. Meanwhile, OpenAI has released a ChatGPT desktop app for Mac users, offering enhanced integration features.

• Biotech startup EvolutionaryScale has introduced ESM3, an AI model that designs and simulates new proteins, mimicking millions of years of natural evolution. ESM3, with 98 billion parameters and trained on 1.8 trillion tokens of biological data, created a fluorescent protein significantly different from existing ones. Founded by former META researchers, EvolutionaryScale has raised $142 million in seed funding from Amazon and Nvidia.

Quickfire News

• Character AI published new research on optimizing inference, revealing techniques enabling the platform to serve 20,000 AI queries per second — 20% of Google Search’s volume.

• Universal Music Group is partnering with AI startup SoundLabs to use voice cloning technology for creating and controlling AI models of their artists' voices for creative applications.

• HeyGen announced a new $60M Series A funding round for its platform that creates realistic AI avatars, with the company now valued at $500M.

• Newly discovered Grok code lists Midjourney, hinting that the rumored partnership may soon include the advanced image generation model within xAI’s chatbot.

• Butterflies launched to the public, a social media app that allows humans and AI personas to interact through posts, comments, and DMs.

• Poolside secured $400M in new funding for its AI-powered software development tools, bringing the company’s valuation to $2B.

• A London film screening was cancelled following backlash after “The Last Screenwriter” credited its screenplay writing to ChatGPT.

• Y Combinator joined over 140 AI startup founders in signing a letter opposing California’s Senate Bill 1047 on AI regulation, arguing it harms the ability to retain talent and stifles innovation.

• Formula 1 and AWS unveiled ‘Statbot’, an AI-powered system that provides real-time context, trivia, and predictions for personalized race broadcasts.

• OpenAI announced the acquisition of Rockset, with the search and database analytics startup’s tech set to boost retrieval infrastructure for enterprise products.

• Axel Peytavin demonstrated a small, autonomous robot prototype powered by GPT-4o that can see, think, and act based on prompts, with capabilities like picking up trash and shaking hands.

• OpenAI CTO Mira Murati said in an interview that the next generation of GPT will be ‘PhD-level’ compared to GPT-3 (toddler) and GPT-4 (high school), and will be released in a year and a half.

• Unitree posted new demos of its Go2 robotic dog performing handstands, flips, and withstanding and adjusting to force, showing extreme precision and agility.

• TikTok parent company ByteDance is reportedly collaborating with U.S. chip designer Broadcom on an AI chip, aiming to secure high-end processors amid ongoing US-China tech tensions.

• Pietro Schirano unveiled Claude Engineer, an open-source tool enabling 3.5 Sonnet to interact with local files and create and edit code through a chat interface.

• Google announced that students will now gain access to its Gemini AI assistant through school accounts, also introducing a series of new AI-powered tools for educators.

• Amazon’s AI upgrades to its voice assistant will reportedly be called ‘Remarkable Alexa’, with an expected $5-10 monthly fee for access.

• Emergence emerged from stealth with $97.2M in funding for its AI agent platform, also open-sourcing its ‘orchestrator’ tool for optimizing model selection during AI workflows.

• Figma introduced Figma AI, featuring Visual and Asset Search, AI text tools, image generation, and quick prototyping.

• YouTube is negotiating with record labels to license songs for AI-powered music generation tools, aiming to launch new features later this year.

• Israel plans to build its first supercomputer, investing $250M in a national AI program to maintain global tech leadership.

• Formation Bio secured $372M in funding for AI-driven drug development, with its valuation now over $1B.

• Opera released the R2 update to its Opera One browser, including AI-powered image generation and recognition, AI Voice Output, and Page Context Mode for summaries and translations during web browsing.

Closing Thoughts

That’s it for us this week.

If you want a version of this newsletter with pictures etc. Be sure to subscribe below:

https://www.blacklynxbrief.com/subscribe